KLAYM AI — GPT DOTS MARKETING
Expert prompt pack · 8 October 2026

Prompt 01 / Audit the brief
============================================================
MISSION & BOUNDARIES
Read the approved marketing brief and the sources I share. Create an onboarding audit with: known facts, assumptions, conflicting information, missing evidence, and the five most important questions.
Use the current product brief as the authority for feature availability and the approved pricing page for pricing. Do not invent answers. Link each factual claim to its source and date.
Deliver a concise audit and a proposed source-of-truth document for my review. Do not change the original files.

INPUT CONTRACT
Company and market: [ ]; commercial objective/baseline/deadline: [ ]; ICP and buying stage: [ ]; source inventory with owners and dates: [ ]; constraints: [ ].
EXPERT METHOD
1. Inventory each source by purpose, owner, date and authority. Separate product facts, customer evidence, hypotheses and historical material.
2. Compare pricing, positioning, feature availability and audience definitions across files. Quote short conflicting passages and identify the decision owner.
3. Assess whether the evidence supports the requested outcome. Distinguish missing facts that block work from assumptions that allow a provisional draft.
4. Prioritize questions by the decision they unlock, not by document order.
DELIVERABLES
Source register: ID | date | owner | authoritative for | conflict | freshness concern.
Decision brief: objective | audience | offer | evidence | constraints | approver.
Gap register: unknown | business consequence | required evidence | owner | blocking/nonblocking.
ACCEPTANCE
Every material fact must be traceable to an accessible source. Do not silently resolve conflicts or treat a linked but unread document as verified. End with READY, READY WITH ASSUMPTIONS or BLOCKED and explain why.

Prompt 02 / The marketing coordinator charter
============================================================
MISSION & BOUNDARIES
You are coordinating marketing operations for [company]. Our audience is [ICP], our offer is [offer], and the next 30-day outcome is [measurable goal].
Use these approved sources: [links/files]. Keep a marketing workboard in [document]. For each task record owner, deadline, source, deliverable link, status, dependency and next action.
You may research, analyze the shared data, draft content, and update this workboard. Publishing content, sending email, changing ad spend or production pages requires a separate explicit approval for the final proposed action.
Create specialist task briefs when useful, but send them the necessary approved context. Do not assume a new task has all our conversation history.
Return the three highest-impact priorities, a first-week plan, and the decisions I need to make. Mark missing access or evidence as blocked. Do not report a task as complete unless the requested output exists and has been checked.

OPERATING CONTRACT
Inputs: [source versions], [team roles/capacity], [budget limits], [tool permissions], [definition of done]. Distinguish actual tool access from proposed access.
EXECUTION METHOD
1. Translate the 30-day objective into measurable outcomes, then a dependency-ordered backlog. Explain which activities directly support the outcome.
2. Prioritize by evidence, expected usefulness, effort and risk. Avoid numerical precision unsupported by data.
3. For each specialist brief include audience, source excerpts/IDs, task boundary, output format, quality criteria and handoff owner. Do not assume shared memory or create unnecessary parallel tasks.
4. Use explicit states: proposed, ready, in progress, blocked, in review, approved, completed. Completion requires an existing deliverable and its agreed check.
5. Before updating an authorized workboard, read its current state and avoid duplicates. On an ambiguous failure, verify the destination before retrying.
OUTPUT
Prioritized backlog, first-week capacity plan, dependency map, approval queue and decision log.
FAILURE HANDLING
Stale evidence, missing permissions or contradictory instructions create a visible blocker with a next action. Do not claim background execution, persistent memory or scheduling unless supported and confirmed. Keep publish/send/spend permissions separate from planning authority.

Prompt 03 / Voice-of-customer research
============================================================
MISSION & BOUNDARIES
Analyze these [number] customer interviews and support notes for [segment]. Use only the shared evidence.
Create a table: job-to-be-done, trigger, pain, desired outcome, objection, exact source, and confidence. Count customers rather than repeated mentions. Separate customer quotations from your interpretation. Anonymize names in the working output.
Recommend three messaging hypotheses and the next questions to ask. Do not claim that this sample represents the entire market. Deliver an insight brief, not a generic persona.

RESEARCH INPUTS
Participant/account IDs, segment, interview date, source text and relevant product context: [ ]. Define the research question: [ ].
METHOD
1. Build a short codebook for trigger, job, desired outcome, anxiety, objection and alternative. Distinguish customer language from interviewer suggestions.
2. Code evidence at participant/account level. Deduplicate the same story across notes and preserve source locations.
3. Compare segments and positive/negative cases. Report sample counts and conflicting evidence rather than smoothing it into a single persona.
4. Identify purchase situations where our verified strengths matter. Separate functional needs, emotional concerns and buying constraints.
OUTPUT
Evidence matrix: participant ID | excerpt | code | interpretation | contradiction | segment.
Three messaging hypotheses: buyer situation | promise | supporting evidence | proof gap | proposed test.
Five disconfirming research questions and an interview recruitment recommendation.
QUALITY GATE
Do not infer market prevalence from mention frequency. Confidence must reflect evidence quality and breadth, not model certainty. Label thinly supported insights. End with what the team can act on now and what requires further research.

Prompt 04 / The weekly content package
============================================================
MISSION & BOUNDARIES
Using the approved customer insight [link] and product brief [link], propose three content angles for [audience]. Each angle must solve a concrete reader problem and identify the evidence needed.
After I select an angle, prepare: one article outline, a founder LinkedIn draft, a distinct company-page draft, and a newsletter section. Adapt each to the channel instead of copying the same text.
Use [voice examples]. Do not invent customer stories, statistics or results. Attach a claims checklist and link each asset to the editorial brief. Save drafts in [location] for review; do not publish.

PRODUCTION INPUTS
Buyer stage: [ ]; reader problem: [ ]; business goal/CTA: [ ]; approved proof: [ ]; format limits: [ ]; reviewer: [ ].
METHOD
1. Compare the three angles by usefulness, originality, proof strength and commercial relevance. State what the reader will learn or do differently.
2. For the chosen angle, define the original contribution: worked example, expert explanation, first-party evidence or practical template. Flag missing input instead of inventing originality.
3. Build an outline that answers a concrete question in every section. Map claims to sources before drafting.
4. Adapt assets: founder perspective for personal LinkedIn, company relevance for the company page, and a useful standalone takeaway for the newsletter. Preserve the same verified facts across formats.
OUTPUT
Editorial brief; full channel drafts with titles/hooks/CTA; asset IDs and source links; claim register; production handoff.
ACCEPTANCE
Check factual consistency, channel length, repetition, voice and CTA fit. Label character counts as estimates unless counted. Separate a draft's readiness from approval. Do not claim the assets were tested or will go viral. Identify one distribution experiment and its business metric.

Prompt 05 / SEO and GEO opportunity brief
============================================================
MISSION & BOUNDARIES
Use the dated search-query export, page inventory and customer-question list attached. Prioritize ten opportunities for [audience] based on buyer relevance, existing evidence, current page coverage and implementation effort.
For each opportunity return: real question, intent, target URL or proposed page, missing information, source evidence, recommended change and measurement plan.
Do not invent search volumes or claim to have crawled pages you did not access. Label estimated effort and hypotheses. Do not promise rankings, AI citations or a visibility score. Produce briefs for review, not live website changes.

EVIDENCE INPUTS
Include canonical URL, query, date range, impressions/clicks where available, customer question, page excerpt and observed AI answer/citation IDs: [ ].
METHOD
1. Audit missing data and distinguish measured queries from synthetic question suggestions. Normalize URLs and identify duplicate or competing pages.
2. Evaluate buyer relevance, explicit answer coverage, evidence quality, freshness and implementation effort separately. Use deterministic calculations for metrics.
3. Recommend update, consolidate, investigate or create. A new page is not the default when an existing page can answer the need.
4. Prioritize ten opportunities using an explained qualitative rubric. Include counterevidence and dependencies.
OUTPUT
Opportunity | buyer task | observed evidence | intent | current URL | gap | proposed fix | proof needed | effort | owner | success metric.
For the top three, provide an editorial/technical brief with acceptance criteria and review requirements.
MEASUREMENT
Separate organic performance, observed AI citations, referral traffic and conversions. Define comparison period and denominators. Do not infer hidden search traces or causal ranking mechanisms. If a page cannot be read, mark coverage unknown. Keep live edits and sitewide changes outside this brief.

Prompt 06 / Campaign launch coordination
============================================================
MISSION & BOUNDARIES
Coordinate [campaign] for [audience], targeting [conversion] by [date]. Use the approved offer and campaign brief. Create an asset matrix covering landing page, email, social, sales support and reporting.
For each asset list owner, deadline, source, dependency, review status and final link. Check that claims, dates and the CTA agree across the drafts. Identify the critical path and the three biggest launch blockers.
Prepare a launch-readiness report. Do not publish, send, or spend money. Ask for a specific approval once the final assets, destination, audience and proposed action are reviewable.

LAUNCH INPUTS
Launch date/timezone: [ ]; authoritative offer/version: [ ]; eligibility/regions: [ ]; asset inventory: [ ]; team capacity: [ ]; tool capabilities: [ ].
METHOD
1. Build the launch backwards from the deadline. Include product readiness, landing page, measurement, copy/design, QA and approvals.
2. Identify the critical dependency chain and owner availability. Mark unknown deadlines rather than inventing commitments.
3. Check price, currency, billing interval, feature scope, eligibility, availability and CTA destinations across every asset.
4. Assign blocker/major/minor severity to discrepancies and propose minimal corrections.
OUTPUT
Launch board: asset ID | owner | dependency | due date | status | version | approval | destination.
Consistency issue log; three largest blockers; ready/blocked recommendation; launch-day monitoring and rollback plan.
APPROVAL CONTRACT
For each external action present exact asset version, audience, destination, timing and spend impact. Approval of a concept is not approval of final distribution. If later authorized, verify execution in the destination and record result IDs. Do not duplicate sends after an uncertain tool response.

Prompt 07 / Paid campaign analysis
============================================================
MISSION & BOUNDARIES
Analyze the attached ad export for [period], [currency] and [attribution convention]. Reconcile totals first. Identify tracking anomalies, conversion lag and low-volume segments.
Return observations, plausible explanations and proposed tests as separate columns. Prioritize at most three tests, with the changed variable, success metric, budget proposal and stop condition. Show the calculations supporting each recommendation.
Do not alter campaigns, audiences, bids or budgets. Missing values must remain missing, not zero.

ANALYTICAL INPUTS
Supply spend, impressions, clicks, conversions, qualified outcomes and revenue where available, plus event definitions, timezone, attribution window and conversion lag: [ ].
METHOD
1. Reconcile totals with the source and identify inconsistent currencies, duplicated rows, tracking breaks and incomplete periods.
2. Calculate CTR, conversion rate and cost per qualified outcome only with valid denominators. Treat zero denominators as undefined.
3. Segment by audience, creative and placement when volume permits. Separate mix changes from within-segment changes.
4. Consider competing explanations for performance: fatigue, landing-page friction, lead quality, tracking and seasonality.
EXPERIMENT SPECIFICATION
For each test define hypothesis, one main variable, control/treatment, eligible audience, primary metric, guardrails, observation window and decision rule. State the baseline and minimum meaningful effect needed for sample planning; do not invent power calculations.
OUTPUT
Reconciled metric table with formulas; evidence versus hypotheses; three experiment cards; unresolved data questions.
QUALITY GATE
Do not equate attributed ROAS with incrementality. If volume is too low, recommend a directional learning test and label its limits. No automatic budget, bid, targeting or creative changes.

Prompt 08 / Lifecycle email brief
============================================================
MISSION & BOUNDARIES
Prepare a three-email onboarding sequence for [segment] based on the approved product brief. Define the trigger, delay, goal, subject, preview text, body and primary CTA for each email.
Use the approved offer and voice. Identify missing screenshots or product instructions. Include a QA list covering recipients, exclusions, links, mobile preview and any unsubscribe requirements.
Save the sequence as drafts for review. Do not import contacts, change automations or send email.

JOURNEY INPUTS
Event schema: [ ]; segment and consent rules: [ ]; existing journeys: [ ]; frequency caps: [ ]; timezone: [ ]; activation event: [ ].
METHOD
1. Define the next valuable user action and eligibility before drafting messages.
2. Specify entry, wait, send-eligible, success, exit and suppression states. Handle duplicate events, delayed events and missing personalization.
3. Define re-entry rules and priority when another campaign targets the same user. Stop onboarding messages if the intended milestone has already occurred.
4. Write one intent and one main CTA per message, with a neutral fallback for missing fields.
OUTPUT
Journey table: trigger | eligibility | exclusion | delay | message | success event | exit | re-entry.
Complete subject, preheader, body and CTA for all three emails.
QA scenarios: already activated, unsubscribed, duplicate trigger, expired offer, missing field and timezone boundary.
MEASUREMENT
Define activation rate and observation window; use a holdout where feasible. Keep clicks separate from incremental product use.
ACCEPTANCE
Flag unavailable event fields and untested links. Do not invent consent, configure automations or claim messages were sent. Identify the CRM owner who must verify implementation.

Prompt 09 / Conversion experiment plan
============================================================
MISSION & BOUNDARIES
Review [URL or supplied page] for [audience] and [conversion]. Use the funnel export and customer feedback attached if available.
For each issue, separate observed evidence from your hypothesis. Prioritize by likely business relevance, confidence and effort. Propose three experiments with a single primary change, success metric, guardrail metric, required tracking and review owner.
Draft copy for the selected experiment using approved claims. Do not fabricate analytics, customer behavior or expected lift. Do not modify the production page.

DIAGNOSTIC INPUTS
Page/version: [ ]; acquisition source: [ ]; funnel event definitions: [ ]; device/channel breakdown: [ ]; user feedback: [ ]; approved claims: [ ].
METHOD
1. Map visitor intent to promise, proof, objection handling and action. Identify mismatches between acquisition messaging and the page.
2. Rule out or flag measurement/technical problems before attributing conversion changes to copy. Mark form behavior, accessibility and speed as untested unless inspected.
3. Separate observed problems from explanations. Consider traffic mix, seasonality and conversion lag.
4. Rank recommendations with qualitative evidence, impact and effort. Do not invent expected percentage uplift.
OUTPUT
Issue register with source evidence and alternative explanations.
Three experiment cards: hypothesis | single main change | control | eligible traffic | primary metric | guardrails | tracking | owner | decision window.
Exact revised copy for the selected experiment and its claim sources.
EVALUATION
Specify baseline and minimum meaningful effect inputs before sample planning. If traffic is inadequate, recommend a usability study or directional test. Define invalidation conditions such as broken tracking or offer changes. Keep deployment behind approval and retain a rollback version.

Prompt 10 / Weekly growth review
============================================================
MISSION & BOUNDARIES
Prepare the weekly growth review from [sources] for [date range and timezone]. State the freshness, currency, event definitions and attribution convention of each dataset. Reconcile totals and report missing data.
Return: goal versus actual, three material changes, supporting calculations, possible explanations, confidence, and three decisions with named owners. Keep correlations separate from causal claims.
For each recommendation specify what we will test next and what evidence would change our mind. Link the output to the workboard. Do not invent results or combine incompatible denominators.

REVIEW INPUTS
Business goals/baselines: [ ]; source exports: [ ]; metric dictionary: [ ]; comparison period: [ ]; workboard: [ ]; decision owners: [ ].
METHOD
1. Check completeness, timestamp, currency, attribution and compatible denominators before combining data. Label partial periods and provisional results.
2. Compare actuals with targets and a comparable prior period. Show absolute and percentage changes with formulas.
3. Investigate the three changes that matter most commercially. Separate facts, plausible causes and untested explanations.
4. Connect recommendations to experiments or corrective work, with capacity and dependencies. Avoid generating new tasks already on the workboard.
OUTPUT
Executive summary of no more than 200 words.
KPI table: definition | target | actual | comparison | delta | source | caveat.
Decision queue: decision | options | evidence | trade-off | owner | deadline | what would reverse it.
Approved/blocked/completed workboard updates as permitted.
ACCEPTANCE
Do not equate correlation with causation or engagement with revenue. Preserve missing data and state how it affects confidence. If a metric is unreliable, recommend fixing measurement before optimizing against it. Never report a recommendation as an executed change.

Prompt 11 / A bounded recurring brief
============================================================
MISSION & BOUNDARIES
For the next four weeks, prepare a marketing brief every Monday at 9:00 AM Europe/Paris from [specific connected sources]. Use a dated export if a source is unavailable and identify its age.
Update [workboard] with completed deliverables, blockers and the three proposed priorities. Deliver the brief in ChatGPT. Escalate only a launch deadline at risk, a contradictory product claim or a decision requiring my approval.
Confirm the saved schedule, timezone, end date, sources and output destination. If the schedule or source monitoring is unsupported, tell me clearly and propose a manual alternative. Do not publish or send anything externally.

SCHEDULE INPUTS
Start date: [ ]; four occurrence dates: [ ]; timezone: Europe/Paris; source identifiers: [ ]; workboard/delivery destination: [ ]; source freshness limits: [ ]; approved permissions: [ ].
SETUP METHOD
1. Confirm whether recurring execution and source access are actually supported. Inspect existing schedules where possible to avoid duplicates.
2. Translate “next four weeks” into explicit dates and show the final occurrence/end condition. Use the named timezone, not a fixed UTC offset across daylight-saving changes.
3. Define the recurring task's input checklist, output contract and failure behavior. A saved schedule does not prove a successful run.
EACH RUN
Report run timestamp, source freshness and any failed reads. Reuse dated exports only with their age visible. Produce KPI changes, completed deliverable links, blockers and three priorities with owners and evidence. Never fill missing data from memory.
OUTPUT AND CONFIRMATION
Return schedule ID/status only if actually saved, exact timing, end condition and delivery destination. If unsupported, provide the equivalent manual checklist and do not claim scheduling.
BOUNDARIES
No external publication or messages. Define retry behavior that avoids duplicate workboard entries. Escalate the agreed categories only. After the fourth occurrence, stop; if automatic end dates are unsupported, disclose that limitation and propose a supported setup before activation.

Prompt 12 / The five-minute voice check-in
============================================================
MISSION & BOUNDARIES
Give me a five-minute marketing decision brief. Cover: what changed, what is blocked, and the single decision with the greatest impact this week. Distinguish facts from your recommendation.
After our discussion, summarize the decisions in writing, update [workboard], and list exactly which tasks should continue, pause or change. Confirm any schedule changes separately. Do not treat the end of the call as an instruction to stop every task.

CHECK-IN INPUTS
Latest verified review: [ ]; workboard: [ ]; current goal: [ ]; available decision-maker: [ ]; source freshness: [ ].
BRIEF STRUCTURE
0:00–0:45: goal versus current position, using verified numbers only.
0:45–2:00: the two most material changes and what remains uncertain.
2:00–3:00: the critical blocker, its consequence and owner.
3:00–4:30: one decision, two realistic options, trade-offs and your recommendation.
4:30–5:00: the explicit question requiring my choice and immediate next step.
Keep the script approximately 550–650 spoken words; timing is estimated until delivered. If voice is unavailable, provide a readable script instead of claiming an audio call.
AFTER DISCUSSION
Separate agreed decisions, suggestions and unresolved questions. Record my actual choice without treating silence as approval. Update the authorized workboard only for explicit commitments, preserving owner, deadline and source context.
OUTPUT
Written decision log and continue/pause/change list with task IDs. Mark uncertain ownership or dates for confirmation. Verify any requested schedule change separately and do not imply all background tasks stop when the check-in ends.
ACCEPTANCE
No invented metrics, decisions or completion states. State which evidence could change the recommendation. Keep external execution behind its existing approval rules.